基于分位回归的偏最小二乘算法的应用

Application of PLS Algorithm Based on Quantile Regression

文章通过真实数据,对基于分位回归的最小二乘(PLS)算法进行应用研究。一方面保留原PLS无独立性假定、无数据分布要求、兼顾变量间关系、数值计算结果客观等性质,另一方面利用分位回归不要求样本同质性、不受离群点影响等优势,完成应用研究。结果表明,基于分位回归的PLS算法避免样本异质性和数据离群点带来的困扰,不拘泥于展示数据信息的平均水平,详实展示不同分位数下数据全貌,更好地揭示变量间关系规律。

This paper utilizes real data to make an application research on the partial least square(PLS) algorithm based on quantile regression.On the one hand,the original PLS has maintained such properties as no independent assumption,no data distribution requirement,consideration of the relationship between variables,the objective results of numerical calculation and so on;on the other hand,the application research is completed by taking advantage of the partial regression which does not require sample homogeneity and is not affected by outliers.The results show that PLS algorithm based on quantile regression avoids the perplex caused by sample heterogeneity and data outliers,not limited to displaying the average level of data information,but shows the full view of data under different quantiles in detail,so as to better reveal the law of relationship between variables.